A  Tutorial:    Genome-­‐based  RNA-­‐Seq  Analysis  
Using  the  TUXEDO  Package  
 
                                                                       (updated:  2014-­‐10-­‐21)  
 
The  following  details  the  steps  involved  in:  
• Aligning  RNA-­‐Seq  reads  to  a  genome  using  Tophat  
• Assembling  transcript  structures  from  read  alignments  using  Cufflinks  
• Visualizing  reads  and  transcript  structures  using  IGV  
• Performing  differential  expression  analysis  using  Cuffdiff  
• Expression  analysis  using  CummeRbund  
 
All  required  software  and  data  are  provided  pre-­‐installed  on  a  VirtualBox  image.    
See  companion  ‘Rnaseq_Workshop_VM_installation.pdf’  for  details.    Data  content  
and  environment  configurations  are  described  therein  and  referenced  below.  
 
 
Before  Running:  
 
After  installing  the  VM,  be  sure  to  quickly  update  the  contents  of  the  
rnaseq_workshop_data  directory  by:  
 
     %      cd  rnaseq_workshop_2014/  
 
     %      svn  up  
 
This  way,  you’ll  have  the  latest  content,  including  any  recent  bugfixes.  
 
Data  Content:  
 
This  demo  uses  RNA-­‐Seq  data  corresponding  to  Schizosaccharomyces  pombe  
(fission  yeast),  involving  paired-­‐end  76  base  strand-­‐specific  RNA-­‐Seq  reads  
corresponding    to  four  samples:    Sp_log  (logarithmic  growth),  Sp_plat  (plateau  
phase),  Sp_hs  (heat  shock),  and  Sp_ds  (diauxic  shift).  
 
There  are  ‘left.fq’  and  ‘right.fq’  FASTQ  formatted  Illlumina  read  files  for  each  of  the  
four  samples.    Also  included  is  a  ‘genome.fa’  file  corresponding  to  a  genome  
sequence,  and  annotations  for  reference  genes  (‘genes.bed’  or  ‘genes.gff3’).        
 
Note,  although  the  genes,  annotations,  and  reads  represent  genuine  sequence  data,  
they  were  artificially  selected  and  organized  for  use  in  this  tutorial,  so  as  to  provide  
varied  levels  of  expression  in  a  very  small  data  set,  which  could  be  processed  and  
analyzed  within  an  approximately  one  hour  time  session  and  with  minimal  computing  
resources.  
 
Automated  and  Interactive  Execution  of  Activities    
 
To  avoid  having  to  cut/paste  the  numerous  commands  shown  below  into  a  unix  
terminal,  the  VM  includes  a  script  ‘runTrinityDemo.pl’  that  enables  you  to  run  each  
of  the  steps  interactively.    To  begin,  simply  run:  
 
   %    ./runTuxedoDemo.pl    
 
 
The  protocol  followed  below  is  that  described  in    
 
Trapnell  C,  Roberts  A,  Goff  L,  Pertea  G,  Kim  D,  Kelley  DR,  Pimentel  H,  Salzberg  SL,  Rinn  JL,  
Pachter  L.    Differential  gene  and  transcript  expression  analysis  of  RNA-­‐seq  experiments  
with  TopHat  and  Cufflinks.  Nat  Protoc.  2012  Mar  1;7(3):562-­‐78.  doi:  
10.1038/nprot.2012.016.  
http://www.nature.com/nprot/journal/v7/n3/full/nprot.2012.016.html  
 
as  illustrated  below:  
 
 

 
 
  
 
   
Use  Tophat  and  Cufflinks  to  align  reads  and  assemble  transcripts  
 
First,  prepare  the  ‘genome.fa’  file  for  tophat  alignment:  
 
%  bowtie-­‐build  genome.fa  genome  
 
(a)  Align  reads  and  assemble  transcripts  for  sample:  Sp_ds:  
 
Align  reads  using  tophat:  
 
%  tophat  -­‐I  1000  -­‐i  20  -­‐-­‐bowtie1  -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  tophat.Sp_ds.dir  
genome  Sp_ds.left.fq  Sp_ds.right.fq  
 
Rename  the  alignment  (bam)  output  file  according  to  this  sample  name:  
 
%  mv  tophat.Sp_ds.dir/accepted_hits.bam  tophat.Sp_ds.dir/Sp_ds.bam  
 
Index  this  bam  file  for  later  viewing  using  IGV:  
 
%  samtools  index  tophat.Sp_ds.dir/Sp_ds.bam  
 
Reconstruct  transcripts  for  this  sample  using  Cufflinks:  
 
%  cufflinks  -­‐-­‐overlap-­‐radius  1  -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  cufflinks.Sp_ds.dir  
tophat.Sp_ds.dir/Sp_ds.bam  
 
Rename  the  cufflinks  transcript  structure  output  file  according  to  this  sample:  
 
%  mv  cufflinks.Sp_ds.dir/transcripts.gtf  cufflinks.Sp_ds.dir/Sp_ds.transcripts.gtf  
 
Now,  you’re  done  with  running  Tuxedo  on  this  sample.    You  now  need  to  repeat  
these  operations  for  each  of  the  three  other  samples,  as  below:  
 
(b)  Align  reads  and  assemble  transcripts  for  sample:  Sp_hs:  
 
%  tophat  -­‐I  1000  -­‐i  20  -­‐-­‐bowtie1  -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  tophat.Sp_hs.dir  
genome  Sp_hs.left.fq  Sp_hs.right.fq  
 
%  mv  tophat.Sp_hs.dir/accepted_hits.bam  tophat.Sp_hs.dir/Sp_hs.bam  
 
%  samtools  index  tophat.Sp_hs.dir/Sp_hs.bam  
 
%  cufflinks  -­‐-­‐overlap-­‐radius  1    -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  cufflinks.Sp_hs.dir  
tophat.Sp_hs.dir/Sp_hs.bam  
 
%  mv  cufflinks.Sp_hs.dir/transcripts.gtf  cufflinks.Sp_hs.dir/Sp_hs.transcripts.gtf  
  
 
(c)  Align  reads  and  assemble  transcripts  for  sample:  Sp_log:  
 
%  tophat  -­‐I  1000  -­‐i  20  -­‐-­‐bowtie1  -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  tophat.Sp_log.dir  
genome  Sp_log.left.fq  Sp_log.right.fq  
 
%  mv  tophat.Sp_log.dir/accepted_hits.bam  tophat.Sp_log.dir/Sp_log.bam  
 
%  samtools  index  tophat.Sp_log.dir/Sp_log.bam  
 
%  cufflinks  -­‐-­‐overlap-­‐radius  1    -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  cufflinks.Sp_log.dir  
tophat.Sp_log.dir/Sp_log.bam  
 
%  mv  cufflinks.Sp_log.dir/transcripts.gtf  cufflinks.Sp_log.dir/Sp_log.transcripts.gtf  
 
 
(d)  Align  reads  and  assemble  transcripts  for  sample:  Sp_plat:  
 
%  tophat  -­‐I  1000  -­‐i  20  -­‐-­‐bowtie1  -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  tophat.Sp_plat.dir  
genome  Sp_plat.left.fq  Sp_plat.right.fq  
 
%    mv  tophat.Sp_plat.dir/accepted_hits.bam  tophat.Sp_plat.dir/Sp_plat.bam  
 
%  samtools  index  tophat.Sp_plat.dir/Sp_plat.bam  
 
%  cufflinks  -­‐-­‐overlap-­‐radius  1    -­‐-­‐library-­‐type  fr-­‐firststrand  -­‐o  cufflinks.Sp_plat.dir  
tophat.Sp_plat.dir/Sp_plat.bam  
 
%  mv  cufflinks.Sp_plat.dir/transcripts.gtf  cufflinks.Sp_plat.dir/Sp_plat.transcripts.gtf  
 
 
 
Merge  separately  assembled  transcript  structures  into  a  cohesive  set:  
 
First,  create  a  file  that  lists  the  names  of  the  files  containing  the  separately  
reconstructed  transcripts,  which  can  be  done  like  so,  writing  each  of  the  four  
Cufflinks  transcript  GTF  files  to  a  newly  created  ‘assemblies.txt’  file.  
 
%  echo  cufflinks.Sp_ds.dir/Sp_ds.transcripts.gtf  >>  assemblies.txt  
 
%  echo  cufflinks.Sp_hs.dir/Sp_hs.transcripts.gtf  >>  assemblies.txt  
 
%  echo  cufflinks.Sp_log.dir/Sp_log.transcripts.gtf  >>  assemblies.txt  
 
%  echo  cufflinks.Sp_plat.dir/Sp_plat.transcripts.gtf  >>  assemblies.txt  
  
#  verify  that  this  file  now  contains  both  filenames:  
 
%  cat  assemblies.txt    
 
cufflinks.Sp_ds.dir/Sp_ds.transcripts.gtf  
cufflinks.Sp_hs.dir/Sp_hs.transcripts.gtf  
cufflinks.Sp_log.dir/Sp_log.transcripts.gtf  
cufflinks.Sp_plat.dir/Sp_plat.transcripts.gtf  
 
And  now  we’re  ready  to  merge  the  transcripts  using  cuffmerge:  
 
%  cuffmerge  -­‐s  genome.fa  assemblies.txt  
 
The  merged  set  of  transcripts  should  now  exist  as  file  “merged_asm/merged.gtf’.  
 
View  the  reconstructed  transcripts  and  the  tophat  alignments  in  IGV    
 
%  java  -­‐Xmx2G  -­‐jar  /Users/bhaas/IGV/current//igv.jar  -­‐g  `pwd`/genome.fa  
`pwd`/merged_asm/merged.gtf,`pwd`/genes.bed,`pwd`/tophat.Sp_ds.dir/Sp_ds.bam
,`pwd`/tophat.Sp_hs.dir/Sp_hs.bam,`pwd`/tophat.Sp_log.dir/Sp_log.bam,`pwd`/toph
at.Sp_plat.dir/Sp_plat.bam  
 
(Note,  you  may  need  to  resize  the  individual  alignment  patterns  in  the  viewer  by  
dragging  the  panel  boundaries.    Afterwards,  go  to  menu  “Tracks”  -­‐>  “Fit  Data  To  
Window”  to  re-­‐space  the  contents  of  the  viewer)  
  
 
 
 
 
Pan  the  genome,  examine  the  alignments,  known  genes  and  reconstructed  genes.  
 
Do  the  alignments  agree  with  the  known  gene  structures  (ex.  intron  placements)?  
 
Do  the  cufflinks-­‐reconstructed  transcripts  well  represent  the  alignments?  
 
Do  the  cufflinks-­‐reconstructed  transcripts  match  the  structures  of  the  known  
transcripts?    
Identify  differentially  expressed  transcripts  using  Cuffdiff:  
 
%  cuffdiff  -­‐-­‐library-­‐type  fr-­‐firststrand    -­‐o  diff_out  -­‐b  genome.fa    
-­‐L  Sp_ds,Sp_hs,Sp_log,Sp_plat  -­‐u  merged_asm/merged.gtf  
tophat.Sp_ds.dir/Sp_ds.bam  tophat.Sp_hs.dir/Sp_hs.bam  
tophat.Sp_log.dir/Sp_log.bam  tophat.Sp_plat.dir/Sp_plat.bam  
 
 
Examine  the  output  files  generated  in  the  diff_out/  directory.  
 
A  table  containing  the  results  from  the  gene-­‐level  differential  expression  analysis  
can  be  found  as  ‘diff_out/gene_exp.diff’.    Examine  the  top  lines  of  this  file  like  so:  
 
%  head  diff_out/gene_exp.diff  
 
 
Study  transcript  expression  and  analyze  DE  using  CummeRbund:  
 
Use  ‘cummeRbund’  to  analyze  the  results  from  cuffdiff:  
 
%  R  
(note,  to  exit  R,  type  cntrl-­‐D,  or  type  “q()”  ).  
 
#  load  the  cummerbund  library  into  the  R  session  
>  library(cummeRbund)  
 
#  import  the  cuffdiff  results    
>cuff  =  readCufflinks('diff_out')  
 
   
#  examine  the  distribution  of  expression  values  for  the  reconstructed  transcripts  
>csDensity(genes(cuff))  
 

 
 
 
 
**(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  
window  in  order  for  the  plot  to  render  correctly.    This  is  a  bug  
with  the  version  of  R  being  used.)    
#  Examine  transcript  expression  values  in  a  scatter  plot  
Expression  values  are  typically  log-­‐normally  distributed.    This  is  just  a  sanity  check.  
 
>csScatter(genes(cuff),  ‘Sp_log’,  ‘Sp_plat’)  
 

 
 
 
Strongly  differentially  expressed  transcripts  should  fall  far  from  the  linear  
regression  line.  
 
(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  
window  in  order  for  the  plot  to  render  correctly.    This  is  a  bug  
with  the  version  of  R  being  used.)  
 
   
#  Examine  individual  sample  distributions  of  gene  expression  values  and  the  
pairwise  scatterplots  together  in  a  single  plot.  
 
>  csScatterMatrix(genes(cuff))  
 

 
 
 
 
 
(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  
window  in  order  for  the  plot  to  render  correctly.    This  is  a  bug  
with  the  version  of  R  being  used.)    
#  Volcano  plots  are  useful  for  identifying  genes  most  significantly  differentially  
expressed.  
 
>  csVolcanoMatrix(genes(cuff))  
 

 
 
 
(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  
window  in  order  for  the  plot  to  render  correctly.    This  is  a  bug  
with  the  version  of  R  being  used.)    
##  Extract  the  ‘genes’  that  are  significantly  differentially  expressed  (red  points  
above)  
 
#  retrieve  the  gene-­‐level  differential  expression  data    
>  gene_diff_data  =  diffData(genes(cuff))  
 
#  how  many  ‘genes’  are  there?  
>  nrow(gene_diff_data)  
 
#  from  the  gene-­‐level  differential  expression  data,  extract  those  that    
#  are  labeled  as  significantly  different.  
#  note,  normally  just  set  criteria  as  “significant=’yes’”,  but  we’re  adding  an    
#    additional  p_value  filter  just  to  capture  some  additional  transcripts  for    
#  demonstration  purposes  only.    This  simulated  data  is  overly  sparse  and  actually  
#  suboptimal  for  this  demonstration  (in  hindsight).  
 
>sig_gene_data  =  subset(gene_diff_data,(significant=='yes'  |  p_value  <  0.01))  
 
 
#  how  many  genes  are  significantly  DE  according  to  these  criteria?  
>  nrow(sig_gene_data)  
 
 
#  Examine  the  entries  at  the  top  of  the  unsorted  data  table:  
 
> head(sig_gene_data)
gene_id sample_1 sample_2 status value_1 value_2 log2_fold_change
56 XLOC_000056 Sp_ds Sp_hs OK 33560.500 117.6900 -8.15563
136 XLOC_000136 Sp_ds Sp_hs OK 30094.800 246.0650 -6.93433
146 XLOC_000146 Sp_ds Sp_hs OK 1125.700 70.9957 -3.98694
269 XLOC_000056 Sp_ds Sp_log OK 33560.500 108.8130 -8.26877
315 XLOC_000102 Sp_ds Sp_log OK 0.000 440.3590 Inf
336 XLOC_000123 Sp_ds Sp_log OK 753.187 0.0000 -Inf
test_stat p_value q_value significant
56 -4.86421 0.00360 0.257929 no
136 -4.49008 0.00795 0.291731 no
146 -3.99117 0.00640 0.291731 no
269 -4.76096 0.00450 0.291731 no
315 NA 0.00160 0.143550 no
336 NA 0.00005 0.008700 yes  
 
#  You  can  write  the  list  of  significantly  differentially  expressed  genes  to  a  file  like  so:  
 
>  write.table(sig_gene_data,  'sig_diff_genes.txt',  sep  =  '\t',  quote  =  F)  
 
   
#  examine  the  expression  values  for  one  of  your  genes  that’s  diff.  expressed:  
 
#  select  expression  info  for  the  one  gene  by  its  gene  identifier:  
#  let’s  take  the  first  gene  identifier  in  our  sig_gene_data  table:  
 
#  first,  get  its  gene_id  
>  ex_gene_id  =  sig_gene_data$gene_id[1]  
 
#  print  its  value  to  the  screen:  
>ex_gene_id  
 
#  get  that  gene  ‘object’  from  cummeRbund  and  assign  it  to  variable  ‘ex_gene’  
>ex_gene  =  getGene(cuff,  ex_gene_id)  
 
#  now  plot  the  expression  values  for  the  gene  under  each  condition  
#    (error  bars  are  only  turned  off  here  because  this  data  set  is  both  simulated    
#      and  hugely  underpowered  to  have  reasonable  confidence  levels)  
 
>  expressionBarplot(ex_gene,  logMode=T,  showErrorbars=F)  

 
 
(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  window  in  order  for  the  
plot  to  render  correctly.    This  is  a  bug  with  the  version  of  R  being  used.)    
##  Draw  a  heatmap  showing  the  differentially  expressed  genes  
 
#  first  retrieve  the  ‘genes’  from  the  ‘cuff’  data  set  by  providing  a    
#    a  list  of  gene  identifiers  like  so:  
>sig_genes  =  getGenes(cuff,  sig_gene_data$gene_id)  
 
#  now  draw  the  heatmap  
>csHeatmap(sig_genes,  cluster='both')  
 

 
 
(Note,  in  the  Ubuntu  VM,  you  may  need  to  resize  the  graphics  window  in  
order  for  the  plot  to  render  correctly.    This  is  a  bug  with  the  version  of  R  
being  used.)  
 
    
More  information  on  using  the  Tuxedo  package  can  be  found  at:  
 
Trapnell  C,  Roberts  A,  Goff  L,  Pertea  G,  Kim  D,  Kelley  DR,  Pimentel  H,  Salzberg  SL,  
Rinn  JL,  Pachter  L.  
Differential  gene  and  transcript  expression  analysis  of  RNA-­‐seq  experiments  with  
TopHat  and  Cufflinks.  Nat  Protoc.  2012  Mar  1;7(3):562-­‐78.  doi:  
10.1038/nprot.2012.016.  
http://www.nature.com/nprot/journal/v7/n3/full/nprot.2012.016.html  
 
 
The  CummeRbund  manual:  
http://compbio.mit.edu/cummeRbund/manual_2_0.html  
 
(note,  most  of  the  tutorial  provided  here  is  based  on  the  above  two  resources)  
 
and  the  Tuxedo  tool  websites:  
 
TopHat:  http://tophat.cbcb.umd.edu/  
Cufflinks:  http://cufflinks.cbcb.umd.edu/  
CummeRbund:  http://compbio.mit.edu/cummeRbund/